Fractal Geometry Segmentation of Polarimetric Synthetic Aperture Radar Images
Abstract
This thesis describes how fractal geometry was used to segment synthetic aperture radar (SAR) images. The fractal dimension of each non- overlapping 8x8 pixel region in a 512x512 pixel image was estimated by use of a multidimensional extension of the box counting algorithm. The differences in estimated fractal dimensions were used to differentiate between regions. The SAR images used were represented by quantized surfaces. These surfaces were defined by nine different datasets of various Euclidean dimensions ranging from three to five. The degrees of freedom (Euclidean dimensions) of the datasets were the X and Y coordinates of each pixel and various combinations of power return values, magnitude return values and phase return values for each pixel. The SAR return values for four different SAR transmit and receive polarization configurations were determined by polarization synthesis techniques using the Stokes vectors of the transmitted and backscattered waves and the target reflectivity phase matrix. Phase matrix data for three different scenes was provided by the Jet Propulsion Laboratory, Pasadena, CA. Thesis results indicate that the fractal dimension of an image 'surface' provides a useful method to segment SAR images. Comparisons between segmentations made using various datasets indicate that datasets containing the magnitude and phase return values yield more information helpful to image segmentation than do datasets containing only power return values. (rrh)
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1989
- Accession Number
- ADA216278
Entities
People
- Mark A. Maneely
Organizations
- Air Force Institute of Technology